SmolVLA
A compact vision-language-action (VLA) model for robotic manipulation tasks. It can serve as a frozen policy augmented by external recovery mechanisms.
Timeline (last 90 days)
Backtrack was evaluated on the LIBERO and MetaWorld benchmarks, covering three VLA models: SmolVLA, π0.5, and π0-FAST.
On MetaWorld, all twelve controller configurations tested with SmolVLA outperformed the unassisted policy.
The project plans to integrate lightweight VLA models such as SmolVLA and OpenVLA-micro in the next step.
All Facts (3)
On MetaWorld, all twelve controller configurations tested with SmolVLA outperformed the unassisted policy.
50%UnverifiedBacktrack was evaluated on the LIBERO and MetaWorld benchmarks, covering three VLA models: SmolVLA, π0.5, and π0-FAST.
50%UnverifiedThe project plans to integrate lightweight VLA models such as SmolVLA and OpenVLA-micro in the next step.
50%